Deepseek Releases Free Language Model v3 That Runs Well on Common Hardware Configurations

Chinese AI startup DeepSeek has just officially released its latest large language model (LLM), DeepSeek-V3-0324. With a capacity of up to 641GB, this model was announced on the Hugging Face AI platform without much information being revealed, in line with the company's usual secretive style about new products.

What makes this model unique is that it is licensed under the MIT license, which allows it to be used freely for commercial purposes. Initial benchmark results show that DeepSeek-V3-0324 is capable of running on mainstream hardware configurations, such as Apple’s Mac Studio with an M3 Ultra chip. AI scientist Awni Hannun reported that it was possible to achieve a processing speed of more than 20 tokens per second using this configuration. This ability to run a large language model on commodity hardware on-premises is a stark contrast to the traditional approach of using massive data center infrastructure to support advanced AI models.

According to DeepSeek, initial tests have shown significant improvements over previous versions. The model has been rigorously tested by internal stakeholders and has demonstrated excellent performance, even outperforming all competing models and beating Anthropic’s Claude Sonnet 3.5 on tasks that do not require deep learning. However, unlike Sonnet-style models that require a paid subscription, DeepSeek-V3-0324 is completely free to download and use.

Technically, DeepSeek-V3-0324 uses a mixture of experts (MoE) architecture. It is notable for its selective use of approximately 37 billion of the 685 billion parameters for each task, which increases efficiency by reducing computational demands while maintaining performance. The model also applies Multi-Head Latent Attention (MLA) and Multi-Token Prediction (MTP) technologies, which contribute to improved context memory and faster output.

Users can access DeepSeek-V3-0324 via Hugging Face, OpenRouter's chat interface and API, and DeepSeek's chat platform if they wish. Inference service provider Hyperbolic Labs also provides access to the model.

With the release of DeepSeek-V3-0324, the company continues to assert its position in the race to develop large language models, while providing a powerful and accessible option for the AI ​​research and development community. The ability to run on commodity hardware, coupled with a free-to-use license, will certainly open up many opportunities for practical applications for this model in the near future.